AI Has an Inclusivity Problem, Some Creators Are Looking to Change It

While interviewing Beth Coleman, Ph.D., Associate Professor of Data & Cities at the University of Toronto, she issues a challenge: “See if ChatGPT can write this article for you; test it.”  Coleman adds, “I'm telling you right now, it can't.” 

She’s responding to a fear that has circulated quite a bit lately. Could artificial intelligence (AI) render creatives jobless?

Coleman’s words offer relief – even if only temporary, given the rapid pace at which AI is evolving. Text generated by a bot tends to be far more easily detectable than an image, though, necessitating the use of human writers to insert nuance, provide context, and even correct grammar. However, if AI-powered image generators like Bing Image Creator and dream by WOMBO can replicate the work of artists and illustrators, won’t the latter eventually become replaceable?

“It's not a competition between humans and the machines,” says artist Victor Wong, an early adopter of AI who, in 2018, invented the robot A.I. Gemini to create traditional Chinese ink paintings. “It’s a collaboration. Sometimes it's even more than a tool because the artwork will inspire interest in the original creator.”

In an ideal world, that would be the standard outcome. The way AI art generators work is by pulling from existing data – text, images, art, videos, audio—to create new content based on a prompt such as “paint a purple flower in the surrealist style of Salvador Dali.” However, as a January article in The Kansas City Defender on “The Explosion of AI Artwork & Its Harmful Impact on Black Creatives” explained, contemporary artists’ work can also be fed into these programs without their knowledge or permission. And they’re often uncredited for the resulting artificial artwork that mimics their own. 

When Adobe launched its AI generative product Firefly in March, the company noted the platform would only pull from its stock image site, openly licensed work, and public domain content to create new works. On March 16, the U.S. Copyright Office also launched a new initiative to examine copyright law and policy issues raised by AI, “including the scope of copyright in works generated using AI tools and the use of copyrighted materials in AI training,” as described in a press release.

“A lot of governments are working on policies to say you can't just take people's stuff,” says Wendy Chun, Ph.D., Director of the Digital Democracies Institute at Simon Fraser University. “There are some fundamental thoughts around how we should make knowledge accessible and how shouldn't we. 

Just last week, Sam Altman, CEO of OpenAI, the creator of the popular ChatGPT chatbot tool, testified before a Senate Judiciary subcommittee about the risks of unregulated artificial intelligence.

“We think that regulatory intervention by governments will be critical to mitigate the risks of increasingly powerful models,” he said in opening remarks, later suggesting licensing and testing requirements for the “development and release of AI models above a threshold of capabilities.” 

What that threshold would be has yet to be established.

The training data set of generative AI programs has been a hot-button issue from the outset, not only in terms of copyright but also racial bias. When OpenAI introduced its second-generation AI image system DALL-E, an external review found that “DALL-E 2’s depictions of people can be too biased for public consumption,” Wired reported. “Early tests by red team members and OpenAI have shown that DALL-E 2 leans toward generating images of white men by default, overly sexualizes images of women, and reinforces racial stereotypes.”

Biases such as these prompted Tavonia Evans to create Melanated AI, a platform that curates AI artwork from creators of color.

“I used to work in graphic design, and the big problem back then was the lack of diverse images,” she says. “Now, fast forward, we have all of these image generators based off of models, and they pull from what's available,” she explains. “There's already a lack of diverse images out there so you can imagine what they're pulling from to generate images. We don't want to continue to propagate that lack of diversity.”

That issue feeds into the immediate backlash Levi’s received when it announced plans this March to begin testing AI-generated clothing models.The denim brand clarified that the models would benefit consumers by allowing them to see what different articles of clothing look like on diverse body types. Yet despite denying any plans to scale back its use of real models, the question of why AI-generated models would be used for diversity efforts rather than hiring real models that are representative of their customer base remained.

In addition to providing a platform for artists to showcase and sell their work, Evans hosts classes to teach budding AI artists how to use her preferred generative art platform Midjourney. “We still suffer from some of the same issues that people of color suffer from in the traditional world, even in the AI world,” explains Evans. “I want to make sure that we're able to address those things and that there's equity because this is something that's moving at the speed of light.”

Coleman, who’s a visiting senior researcher with Google’s Brain team, as well as their Responsible AI team, which makes sure companies are making ethical considerations when developing generative AI programs, says the solution sits with who’s training the models.

“There is not enough inclusivity or diversity in how most of the models are trained. Because it takes a lot of computing power and also a lot of people working on the projects to put together something like ChatGPT, one of the things that’s been established as a good practice is to have a certain level of diversity in the room in the beginning.”

Coleman will soon release a catalog with Berlin press K. Verlag that shows what’s possible when AI is used to combat stereotypes rather than feed into them. The work is inspired by science fiction writer Octavia Butler and the limitless societies she created with her novels.

“It's called Reality Was Whatever Happened: Octavia Butler AI and Other Possible Worlds and I've got a whole set of images that are based on this concept of what happens when you use machine learning for generative AI to break down silos and boundaries about identity and even humanness,” she says.

While the general sentiment among those who’ve leaned into generative AI is that it should be used to inspire new human creations rather than replicate others’ work, Wong, who also owns award-winning visual effects company, vfxNova, says it has practical benefits for creators in other artistic realms. One of those being marketing and advertising, as the technology allows for quick creation of storyboards and creative deks to present to clients.

“In the past, it would take two weeks to do that preparation. Now, in a few hours, I can come up with preliminary compositions,” he says.

Rapid production is what made Evans fall in love with AI art. As a tech entrepreneur, she doesn’t have time to physically paint a picture, but with generative art, she can still create. “I'm a prompt engineer so I'm imagining what I want to see, and I'm talking to AI to get it to make this art for me,” explains Evans who’s the founder of the cryptocurrency Guapcoin. “I can do that from my phone. I can do it in bed. I can get up at 2 o'clock in the morning and have this really great idea for a piece of art that I want, type in the prompt and get it. I'm like a kid in a candy store.”

As for where generative AI art is headed, Wong says, “I think AI is a river— really, it's a waterfall now. It will wash away artists if you don’t know what you’re doing. But if you put yourself in this river and then you flow with it, you never know [where it could lead you].”

Evans echoes that sentiment, suggesting AI can be more helpful than harmful to artists of color if they use it to their advantage. “I say you better dive in. It's not going anywhere. So if you can't beat it, join it.”

As for making AI generative art more representative, the responsibility may not solely lie with the technology itself but rather the social structures that have laid the foundation for such biases to come to a head.

“It’s not a wokeness or a PC problem,” Coleman emphasizes. “It’s a real problem in terms of the technologies being built into how society works and represents itself.”

Let us slide into your DMs. Sign up for the Teen Vogue daily email.

iPad OS 17’s Live Collaboration on PDFs could challenge Google Docs

Apple knows that iPad users are often interacting with PDF files, whether they’re studying for a chemistry exam or filling out paperwork at the doctor’s office. So, it only makes sense that on iPad OS 17, Apple is adding features that improve the experience of working with PDFs on iPad, as announced at WWDC 23.

With its machine learning technology, iPad OS 17 can identify fields in a PDF so that you can fill them out quickly — that’s big for people who don’t have easy access to Adobe Acrobat. This also functions with photos of documents that you scan onto your iPad.

Image Credits: Apple

Some of the most exciting PDF upgrades come inside of the Notes app. You can access multiple PDFs within one note, and if you’re collaborating with another iPad OS 17 user, you can make edits, Apple Pencil notes or other changes that update immediately on each device. The user experience seems similar to working with a fellow student on a homework assignment in Google Docs, only the functionality with Apple Pencil could be extra useful for annotating documents or taking notes with co-workers or classmates.

This feature, called Live Collaboration, can be used remotely. It also can be used in conjunction with a FaceTime call.